What is the state of machine learning at the edge today? What tools can help engineers collect data and run inferences? Where can you find ST MEMS, and how can they make a difference in real-world products? This piece is the second part of our series on the upcoming STM32 Roadshow. For the 14th year in a row, we are reaching out to our community. We will hold demos, show products, and have engineers ready to answer questions. The first part of our STM32 Roadshow Series focused on cloud connectivity as we talked about a new smart doorbell demo. We also featured industrial and security applications. Today, we will explore artificial intelligence and computing as well as sensing.

如今網絡邊緣側的機器學習現狀如何？哪些工具可以幫助工程師收集數據并執行推斷運算？在哪里可以找到ST MEMS，它們對現實生活中的產品有哪些影響？本文是我們即將舉行的STM32全國研討會系列的第二篇專題文章。在第14屆STM32全國研討會上，我們將通過應用演示、產品展示以及工程師與觀眾互動回答問題的方式，來與蝶粉社區近距離交流。在STM32全國研討會專題系列報道第一部分我們著重介紹了云連接方面的用例，如一款新的智能門鈴功能演示，還介紹了工業和數據安全相關應用。 今天，我們將重點探討人工智能、計算以及感知技術。

Artificial Intelligence and Computing

人工智能與計算

Qeexo 和STM32Cube.AI

The range of solutions enabling machine learning at the edge is also increasing, and the STM32Roadshow highlights the central role of STM32 MCUs. For example, we will show a demo of Qeexo’s AutoML. It uses a SensorTile to capture vibrations and sounds to detect if a fan is broken or blocked. It is a classic example of a predictive maintenance application that can vastly transform a factory’s operation with minimal investments. Qeexo is a member of the ST Partner Program.

當今邊緣機器學習解決方案的種類越來越多，本屆STM32全國研討會將聚焦討論STM32 MCU在這類應用中的核心角色。我們將演示Qeexo’s AutoML工業自動化機器學習解決方案（Qeexo是ST合作伙伴計劃成員）。該系統使用SensorTile捕獲振動和噪聲，檢測風扇是否損壞或阻塞，這是一個經典的，以最少的投資來最大化提高工廠運營效率的預測性維護應用示例。

There will also be numerous ST demos that leverage our machine learning solutions. Some of them are already popular, such as the STM32H747I-DISCO that uses machine learning to recognize dishes and drinks. It was a show favorite at the Technology Tour in Toronto and remains popular amongst our attendees. Our engineers will also demonstrate a system capable of reading a digital meter. This particular presentation uses an STM32WL, our first MCU, with an embedded LoRa transceiver.

全國研討會上還有很多ST的機器學習應用演示，其中一些已經很有人氣，例如，使用機器學習識別食品飲料的STM32H747I-DISCO。它在Technology Tour in Toronto（多倫多科技展）上廣受關注，在本屆全國研討會參觀者中也仍享有很高的人氣。我們的工程師還將演示一個智能電表抄表系統，這個特別的演示使用的是STM32WL——我們的第一款帶有嵌入式LoRa收發器的MCU。

Similarly, the STM32MP1 will run on a new AI demo offering multiple object detection. We rewrote the code in C to optimize it, and it will be the first time we show it in Asia. Moreover, we will showcase FP-AI-NANOEDG1, a Function Pack that allows developers to quickly test a Machine Learning library from Cartesiam on an STM32L5.

同樣，STM32MP1將出現在一個新的AI多物體檢測演示板上。我們重寫并優化了C語言代碼，這個解決方案是首次在亞洲演示。此外，ST還將展示一個使開發人員可以在STM32L5上快速測試Cartesiam機器學習庫的FP-AI-NANOEDG1功能包。

OpenMV

The STM32 Roadshow will be a great place to experience the OpenMV Cam H7 Plus. The product relies on an STM32H7 microcontroller to capture videos using a five-megapixel camera module on top of the PCB. Additionally, the platform works using MicroPython to make it easier to program. It thus puts a robust system in the hands of engineers and enthusiasts wishing to experiment with embedded systems quickly. Users can even download the OpenMV IDE and run example applications that will show some of the system’s capabilities.

本屆STM32全國研討會將是觀眾體驗OpenMV Cam H7 Plus的絕佳機會。該產品依靠STM32H7微控制器和PCB板載500萬像素攝像模塊拍攝視頻。此外，該平臺還可以支持MicroPython語言，使編程變得更輕松，它為那些希望快速測試嵌入式系統AI的工程師和發燒友提供了一個穩健的系統。用戶甚至可以下載OpenMV IDE開發環境，運行系統功能演示應用程序，查看某些系統功能。

The event will also demonstrate to attendees that they can go much further than the typical demos. For instance, Edge Impulse has a tutorial showing how to write a machine learning application with the OpenMV Cam H7 Plus. The ST Partner Program member facilitates the creation of neural networks that can then run inference operations on ST’s MCUs. In this instance, developers use the OpenMV PCB and IDE to collect data. They then send it to Edge Impulse for processing. Finally, users can export a neural network as an OpenMV library. This system is also impressive because as engineers transition to an industrial setting, it is possible to use Edge Impulse to get a neural network that will work with STM32Cube.AI. This software solution converts neural networks into optimized code for STM32 to vastly facilitate machine learning at the edge.

觀眾還將在本屆研討會上了解到比一般demo演示更深層次的東西。例如，Edge Impulse（ST合作伙伴計劃成員之一）有一個如何使用OpenMV Cam H7 Plus編寫機器學習應用程序的教程，讓開發在ST MCU上執行推斷運算的神經網絡變得更容易。在這個示例中，開發人員可以使用OpenMV PCB和IDE收集數據，然后，發送到Edge Impulse進行數據處理，最后，可以導出神經網絡的OpenMV庫。該系統令人印象深刻。隨著工程師開始關注工業環境，使用Edge Impulse就可以獲得一個支持STM32Cube.AI的神經網絡。該軟件解決方案將神經網絡轉換為可在STM32上運行的代碼，從而極大地降低了邊緣機器學習的開發難度。

Sensing and Innovation

感知與創新

SensorTile.box and the Crying Baby Detector

SensorTile.box和寶寶哭聲檢測器

The SensorTile.box will be another highlight of the STM32 Roadshow. Our most powerful sensor box with multiple user modes will be at the center of a few demos. Users will be able to interact with built-in demo applications. The STEVAL-MKSBOX1V1 (the reference of the SensorTile.box) with iOS and Android applications to quickly showcase some of its capabilities. For instance, ST provides a baby crying detector. The application first uses an algorithm that employs a Fast Fourier Transform to process the signal. It then runs the data through a neural network on the host STM32. Thanks in part to STM32Cube.AI, developers can use a regular MCU to distinguish between ambient noise and a child’s cries. This demo is also highly symbolic because it exemplifies how our sensors, MCUs, and more work to create unique and wholesome solutions.

SensorTile.box將是STM32全國研討會的另一個亮點。我們最強的多用戶模式傳感器模組SensorTile.box將是幾場演示活動的核心角色。用戶將能夠與內置的演示應用程序互動，裝有iOS和Android應用程序的STEVAL-MKSBOX1V1（Sensor-Tile.box的型號）可快速展示模組的部分功能，例如，ST提供的寶寶哭聲檢測器。該應用先是運行一個算法，采用快速傅立葉變換方法處理信號；然后，通過主控制器STM32上的神經網絡運行數據。開發人員可以使用常規MCU辨別環境噪聲和孩子的哭聲，其中，STM32Cube.AI功不可沒。該演示還具有高度的示范意義，因為它是一個展示我們的傳感器、MCU等芯片如何協同工作，創建獨特而有益的解決方案的范例。

OPPO Smartwatch and Edifier Dreampods

OPPO智能手表和漫步者耳機

The STM32 Roadshow will also be an opportunity to check out significant design wins physically. For instance, we will showcase an OPPO smartwatch that includes our LPS27HHW barometer. The component can measure how deep a user is swimming or how high that person is climbing. The OPPO watch also includes the LSM6DSOW, which uses finite state machines to detect human activities while reducing the overall power consumption. The system can thus detect if a user is running or cycling while consuming very little to save its battery.

本屆STM32全國研討會的另一個看點是，觀眾將有機會看到幾個重要的ST設計采納用例。舉例來說，我們將展示一個OPPO智能手表，這款手表內置我們的LPS27HHW氣壓計傳感器，可以測量用戶游泳水深或攀爬高度。OPPO手表還集成了LSM6DSOW慣性測量單元，它使用有限狀態機檢測人類活動，同時能夠降低系統總體功耗。因此，該系統可以檢測用戶是在跑步還是騎車，而且幾乎不耗電，十分節省電池電量。

Similarly, we will also showcase the Edifier Dreampods. It is fascinating to learn how these wireless earphones use a LIS25BA to detect vibrations crawling from the inner ear to the facial bones. Such a system ensures the device can distinguish between the audio and ambient noise. The Dreampods also use the LIS2DH12 accelerometer to enable users to tap on the earphones to play or pause music and operate other controls, such as picking up a call or hanging up. Both the Dreampods and the OPPO smartwatch are available on the Chinese market.

同樣，我們還將展示漫步者的Dreampods耳機。了解這些無線耳機如何使用LIS25BA檢測從內耳傳向面部骨骼的振動對開發者抑或耳機發燒友而言無疑是一件非常有趣的事情。該系統確保設備可以區分音頻和環境噪聲。 Dreampods還集成了LIS2DH12加速度計，用戶只要敲擊耳機就可以播放或暫停音樂，還可以進行其他控制操作，例如接聽電話或掛斷電話。現今Dreampods和OPPO智能手表都能在中國市場買到。